/healthcare
Product leadership with deep healthcare expertise — payer–provider alignment, utilization management as evidence, value-based care that compounds, and AI with a human accountable.
When UM is done well, it is almost like a good referee in soccer — it goes unnoticed.
One Christmas several years ago I got a text asking if I had bought something at a trendy clothes store in London. I was in Boston. I said definitely not — hold my card, there is fraud. Later my wife mentioned she had trouble buying presents for family back in the UK because the card was not processing. It clicked. These were good transactions; I just was not aware. I called the bank, reactivated the card, and we re-processed the purchases.
UM in the U.S. needs to become that seamless. When the provider is doing the right things, information flows from their EMR to the payer and approvals can happen automatically because the right things are happening. You can even get to the point of proactive approval for what the patient will need next: you ordered a hip replacement — I have gone ahead and pre-approved 30 days of home care and PT to help the recovery.
The biggest argument is that we have been waiting for VBC for 30 years and it has not happened. Too much care is still fee-for-service. CMS has made real progress moving the industry, and it is hard. Every additional percentage shift from paying for services to paying for value makes the system better for all of us.
I met someone who worked for a health system that was 50% FFS and 50% VBC. They had 64 cardiologists. They figured out who the best four were and had them serve the 50% they were at risk for. The other 60 served the 50% FFS. They put processes in place to elevate those four so they made a massive impact. That is the order of magnitude of outcome when you talk about VBC — and the transition is still hard, because the other 60 are not all needed anymore.
A good example of breaking the problem down: when we used AI to interpret faxes, there were duplicates. Digging in, the team found several faxes were the exact same as something received 2 or 5 or 15 minutes before. Instead of burning tokens to re-process them, a simple checksum could catch exact duplicates — log that it happened and move on. For some clients that cut 25% to 30% of what needed to be processed by AI.
Deterministic work as rules. AI where AI earns its keep. And someone accountable when it drifts.
There need to be integrated, reinforcing checks on the KPIs so the right things are measured. Years ago, when we rolled out auto-approval for prior authorization, one client had three providers suddenly at 100% approvals while their cohort ran closer to 85%–90%. Provider Relations met with each. The first two said, in effect, we figured out how to get approvals — we will be more careful. The third was adamant they were doing it right and their patients all needed the care. That third went onto a manual review list with no more auto approvals.
Watch the process. Understand the outliers. If you own UM, history and expectations matter — and anything that deviates should be understood.
In addition to product leadership, this is where I have spent most of my career proving the structure-and-incentives thesis — payer–provider systems, evidence-based UM, VBC that compounds slowly, and AI in production with a human who can stand behind the decision.